Analysis of water quality at hydrographic basin scale using satellite images, co-occurrence matrices and Bayes classifier

• Published in: Water science & technology. Water supply Vol. 21; no. 8; pp. 4418 - 4428
• Main Authors: Silva, M. G. G., Silva, D. J., Costa, P. D., Silva, R. C., Cassimiro, T. E. B., Amorim, L. S., Rocha, D. A., Peixoto, Z. M. A.
• Format: Journal Article
• Language: English
• Published: London IWA Publishing 01-12-2021
• Subjects: Algorithms; Artificial intelligence; Basins; bayes classifier; Bayesian analysis; Chemical oxygen demand; Classification; co-occurrence matrix; Computer applications; Data collection; Digital imaging; Discriminant analysis; Environmental services industry; Image quality; Internet of Things; Machine learning; Nitrates; Nitrogen; Quality assessment; satellite image; Satellite imagery; Satellites; Saturation; Sensors; Software; Spaceborne remote sensing; Support vector machines; Training; Water analysis; Water pollution; Water quality; Water quality assessments; Water resources; Water scarcity; Brazil; China
• ISSN: 1606-9749; 1607-0798
• DOI: 10.2166/ws.2021.192

Abstract
Given the increased risks of water scarcity and the presence of polluting agents in water resources, this paper aims at the development and presentation of a computational tool capable of assessing water quality based on digital processing techniques applied to satellite images. Initially, a database was created for Brazilian regions, consisting of hydrographic basins' satellite images associated with the Water Quality Index (WQI), according to the criteria established by the National Water Agency (ANA). Hitherto, the database consisted of 85 images, 61 were used in the training stage and 24 in the testing stage. In both stages, the images were subjected to thresholding using Otsu's Method, binarization, linear expansion on saturation, application of a Laplacian filter, extraction of characteristics by using co-occurrence matrices and classification by the Bayes Discriminant. Such techniques were also implemented on a computational platform in the MATLAB® environment, responsible for the interface between the system and users. The proposed system presented an approximate 70% success rate regarding the classification of WQIs, which can be improved as more information is made available to improve the databases.